Reproduction:
- Push a task to try (example job). This fails as expected.
- Edit the revision and push to try again.
- The new decision job fails with error:
abort: revision in set has more than one child. (example job)
- Wait for the worker to time out.
- Push the same try job again. It now runs correctly. (example job)
Example log showing the error. It looks like the workers might not be purging their local checkouts correctly between decision jobs?